The Intervention Gap in Latent World Models

• URL: http://arxiv.org/abs/2608.29998v1
• Date: Sun, 30 Aug 2026 19:47:51 GMT
• Title: The Intervention Gap in Latent World Models
• Abstract summary: Planning-time intervention fidelity is a distinct, measurable property of a learned world model.<n>Across released TD-MPC2 checkpoint sizes, episode return falls as an operator-error diagnostic on task observables grows.<n>A self-supervised world model trained without task signal preserves the same operator substantially better than a task-anchored model on the shared task.
• Score: 0.0
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Planning-time intervention fidelity is a distinct, measurable property of a learned world model: whether the model's own open-loop transitions move task variables the way matched environment interventions do. In the settings we test, it is neither revealed by reward fit nor ensured by task-anchored training. Across released TD-MPC2 checkpoint sizes, episode return falls as an operator-error diagnostic on task observables grows, while reward-prediction error stays small and nearly flat, and a self-supervised world model trained without task signal preserves the same operator substantially better than a task-anchored model on the shared task. A capture-gated matched-intervention audit then localizes what fails. On Cheetah, three LeWorldModel checkpoints capture the current task query and support decodable real intervention effects; however, their imagined five-step effects are worse than predicting no effect and worse than an environment-endpoint oracle. The failure is task-direction rotation with excess gain, not feature collapse. This severe pattern is conditional: five PreJEPA seeds retain an oracle-relative deficit without it, Finger Spin experiments extend the deficit beyond locomotion with heterogeneous severity across seeds, and shared-bank effect geometry is both candidate- and support-dependent. We also test practice-side questions. In DreamerV3 the posterior distribution, not its sample, carries the current query; ensemble disagreement ranks error only near training support; and a frozen support-aware score degrades held-out error ranking in both tested transfer directions while native disagreement remains informative in both. We conclude that intervention fidelity must be audited directly, capture-first, on the model's native interface.

Related papers

• Vision: Data-Centric Anchoring for Robust and Interpretable Agentic AI [21.406298170489507]
  We argue that robustness and interpretability should be engineered into the data environment, not extracted from models after training.<n>Our central contribution is the Data-Centric Agentic Loop, a four-stage framework of Curate, Augment, Constrain, and Attribute.<n>We ground the framework in a failure-driven taxonomy that links four core failure modes to the data lifecycle.
  arXiv  Detail & Related papers  (2026-09-08T03:57:54Z)
• ObserverBench: Testing Mechanistic Estimates for Intervention and Control [0.0]
  Mechanistic interpretability is increasingly used to guide interventions such as activation steering, circuit removal, and safety monitoring.<n>We present ObserverBench, a benchmark framework for testing whether an internal estimator is adequate for the intervention, control, or safety task it directs.<n>The benchmark reports estimation accuracy separately from the loss caused by the chosen action.
  arXiv  Detail & Related papers  (2026-09-02T18:01:17Z)
• Suppression Sticks, Locality Is Fragile: A Closed-Loop Target-and-Control Audit of Task-Vector Negation in VLA Policies [33.432377306905735]
  Task-vector arithmetic offers a closed-form way to modify a model, yet its behavioral locality remains unclear in closed-loop robot control.<n>We present a target-and-control audit of per-skill task-vector subtraction from multitask vision-language-action (VLA) policies.
  arXiv  Detail & Related papers  (2026-08-05T10:59:53Z)
• The Knowing-Saying Gap: When Probes See Errors that Confidence Misses [20.46211358172975]
  Linear probes detect corrupted context in language models with near-perfect accuracy, yet this does not translate into reliable failure prediction.<n>Across multi-hop arithmetic chains, probes that detect corruption turn out to be uninformative about final answer correctness.<n>As a real-time monitor, probe-based interventions are sharply model and error-type dependent.
  arXiv  Detail & Related papers  (2026-07-21T12:13:10Z)
• A Control Theory of Predictability in Latent World Models [39.22646915328561]
  Current practice adopts the prediction error, the single- or multi-step rollout loss on held-out data, as the training and model-selection objective.<n>We show that this assumption is unreliable for a structural reason.<n>A planner does not query the model on the training distribution but on the states that its candidate actions reach, which generally leave the data manifold.
  arXiv  Detail & Related papers  (2026-07-11T15:42:26Z)
• Survive or Collapse: The Asymmetric Roles of Data Gating and Reward Grounding in Self-Play RL [76.45061154544568]
  Self-play reinforcement learning trains language models on their own generated tasks, co-evolving a proposer and solver without human labels.<n>Recent systems report strong reasoning gains, but collapse and instability are widely observed and poorly understood.<n>We argue instead that self-play stability is governed by two distinct levers: a data-level gate that decides which proposer-generated tasks enter the training pool, and the reward signal that updates the policy on tasks already admitted.
  arXiv  Detail & Related papers  (2026-05-21T09:19:23Z)
• On the Paradoxical Interference between Instruction-Following and Task Solving [50.75960598434753]
  Instruction following aims to align Large Language Models (LLMs) with human intent by specifying explicit constraints on how tasks should be performed.<n>We reveal a counterintuitive phenomenon: instruction following can paradoxically interfere with LLMs' task-solving capability.<n>We propose a metric, SUSTAINSCORE, to quantify the interference of instruction following with task solving.
  arXiv  Detail & Related papers  (2026-01-29T17:48:56Z)
• Reliably Detecting Model Failures in Deployment Without Labels [14.069153343960734]
  This paper formalizes and addresses the problem of post-deployment deterioration (PDD) monitoring.<n>We propose D3M, a practical and efficient monitoring algorithm based on the disagreement of predictive models.<n> Empirical results on both standard benchmark and a real-world large-scale internal medicine dataset demonstrate the effectiveness of the framework.
  arXiv  Detail & Related papers  (2025-06-05T13:56:18Z)
• Collaborative Value Function Estimation Under Model Mismatch: A Federated Temporal Difference Analysis [55.13545823385091]
  Federated reinforcement learning (FedRL) enables collaborative learning while preserving data privacy by preventing direct data exchange between agents.<n>In real-world applications, each agent may experience slightly different transition dynamics, leading to inherent model mismatches.<n>We show that even moderate levels of information sharing significantly mitigate environment-specific errors.
  arXiv  Detail & Related papers  (2025-03-21T18:06:28Z)
• Nonparametric Identifiability of Causal Representations from Unknown Interventions [63.1354734978244]
  We study causal representation learning, the task of inferring latent causal variables and their causal relations from mixtures of the variables. Our goal is to identify both the ground truth latents and their causal graph up to a set of ambiguities which we show to be irresolvable from interventional data.
  arXiv  Detail & Related papers  (2023-06-01T10:51:58Z)
• Tracking the risk of a deployed model and detecting harmful distribution shifts [105.27463615756733]
  In practice, it may make sense to ignore benign shifts, under which the performance of a deployed model does not degrade substantially. We argue that a sensible method for firing off a warning has to both (a) detect harmful shifts while ignoring benign ones, and (b) allow continuous monitoring of model performance without increasing the false alarm rate.
  arXiv  Detail & Related papers  (2021-10-12T17:21:41Z)
• Nested Counterfactual Identification from Arbitrary Surrogate Experiments [95.48089725859298]
  We study the identification of nested counterfactuals from an arbitrary combination of observations and experiments. Specifically, we prove the counterfactual unnesting theorem (CUT), which allows one to map arbitrary nested counterfactuals to unnested ones.
  arXiv  Detail & Related papers  (2021-07-07T12:51:04Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.